Key priorities for the EIOS system

For the EIOS system, a number of priorities can be identified along the three lines of action discussed in the sections that follow. These primary issues are consistent with those that were raised by stakeholders in the EIOS Global Technical Meetings held in Geneva (2018) and Seoul (2019).

The developments envisaged will be carried out in the framework of existing and future agreements between the JRC, the WHO and other EIOS stakeholders and collaborators, including external partners such as universities and research institutes.

7.2.1 Improving early detection and early warning

Early detection can be improved by incorporating even more potentially relevant information, and by increasing the ability of the system to automatically extract knowledge from the information and therefore highlight potential sources of threats.

Along these lines, priorities include the following:

— Monitor social media (dashboard with information from Twitter and incorporation of this information in the EIOS system), including automated sentiment analysis

— Monitor misinformation (installation of tools already developed by the JRC and subsequent integration in the EIOS system) to cope with ‘infodemics’

(12_{) https://www.biorxiv.org/}

36

— Include of further metadata provided by EMM/MEDISYS (14_{): clusters (i.e. grouping articles into stories),} event metadata, entities (e.g. specific people or organisations mentioned in the articles), quotations, and an improved geolocation algorithm; quotes can be extracted automatically and linked to entities. Other advanced algorithms could be explored to improve the automatic extraction of meaning from articles. — Introduce additional advanced filtering capabilities in the user interface to better define the

monitoring and analysis scope and deal with increasing amounts of information.

— Develop anomaly detection algorithms based on the volume of information on a given topic (e.g. sudden increase in news about a given disease in a given country); a statistical analysis of the volume of information is available from EMM/MEDISYS. It is however important to stress that an increase in information in the news may not necessarily be relevant to public health, but rather may be based on what the media perceives to be of interest to its readership. Thus, it is important to define which anomalies should be captured with specific filters (see previous point).

In addition to the previous proposals, other solutions to enhance early warning could be explored, based on alternative sources of information (such speech-to-text recognition from radio stations) and models.

With regard to the latter, an alternative approach at least for certain diseases could be the employment of ‘omic technologies’ (see for instance Schneider and Orchard, 2011) and genomics in particular. The analysis of mutations in full genomic viral sequences – published online in databases such as GISAID (15_{) – could be} employed to perform ‘genomic surveillance’, that is, monitoring the temporal and spatial distribution of specific mutations associated with increased virulence and infectivity and therefore having a selective advantage. This could be particularly helpful for viruses that have a relatively low sequence mutation rate, such as SARS-CoV- 2 (Romano et al., 2020), and coronaviruses in general (Denison et al., 2011).

7.2.2 Enhancing collaboration and knowledge management

A priority for the EIOS system is enhancing communication and knowledge sharing among communities of practices collaborating on the EIOS platform, with specific tools for reporting and collaborative monitoring and risk assessment.

Another priority is continuing the integration of contextual information to facilitate rapid risk assessments and build specific analytics tools that are able to extract knowledge from these indicators.

Possible development actions include the following:

— Integrate and develop taxonomies and ontologies in the health domain (diseases, symptoms, etc.). Such knowledge can serve as both a form of contextual information and additional input for AI algorithms to improve automatic early detection and analysis.

— Using the existing INFORM database as a start, further develop and Incorporate a library of vulnerability and coping capacity indicators, that includes detailed demographic profiles of the country and specific sub-national, comorbidity index, connectivity and traffic, and mobility and migration data, number of hospital beds and intensive care units.

— Integrate other databases including information about past events, namely data about past outbreaks (epidemic curves and trends in fatalities, hospitalisation, etc.), the specific containment measures adopted to address them (including the need for international aid) and the related impacts. Specific Information regarding the different contingency plans could also be extracted via media monitoring to help analysts understand what has been successful or not in the management of the crisis. This is a necessary step for the development of analytics and predictive capacities to support crisis monitoring and management, addressing the development of the disease in similar scenarios (e.g. countries with comparable profiles and situation), including with the support of AI algorithms.

— The application of AI analysis to previous evaluations of similar threats carried out by analysts on the platform will allow to patterns in diseases to be identified, providing an initial, automatic assessment of potential impacts and the evolution in time and space of the disease.

(14_{) https://medisys.newsbrief.eu/}

37

7.2.3 Expansion to include other stakeholders

The EIOS initiative aims at to expand to include other institutions and communities, such as more WHO member states, non-governmental organisations and networks of public health practitioners.

Beside the need for better supporting collaboration, this long-term strategy poses challenges, including in terms of the technical capacity of the system, with a constant need to monitor and revise the effectiveness of its software architecture.